varuintarray 1.2.0b1

A variable-length unsigned integer array

VarUIntArray

A NumPy subclass for working with variable-length unsigned integers that don't fit standard machine word sizes.

Overview

VarUIntArray extends numpy.ndarray to handle arbitrary bit-width unsigned integers (e.g., 3-bit, 10-bit, 12-bit) while correctly managing padding bits when using NumPy's universal functions (ufuncs). This is particularly useful when working with:

• Custom binary formats with non-standard word sizes
• Packed bit arrays where words don't align to 8, 16, 32, or 64 bits
• Data structures that require precise bit-width control

Key Features

• Arbitrary Word Sizes: Support for any word size from 1 to 64 bits
• Automatic Padding Management: Correctly handles padding bits in bitwise operations
• Byte Order Control: Specify 'big', 'little', or 'native' byte order
• NumPy Integration: Works seamlessly with NumPy ufuncs and array operations
• Pack/Unpack Operations: Convert between bit arrays and packed integer arrays

Installation

This module can be installed from PyPi:

pip install varuintarray

Quick Start

Create a VarUIntArray with 10-bit words

>>> arr = VarUIntArray([1, 2, 1023], word_size=10, byteorder="big")
>>> arr
VarUIntArray([   1,    2, 1023], dtype='>u2', word_size=10)

Bitwise operations respect word_size

>>> inverted = arr.invert()
>>> inverted
VarUIntArray([1022, 1021,    0], dtype='>u2', word_size=10)

Unpack to individual bits

>>> bits = arr.unpackbits()
>>> bits.shape
(3, 10)

Pack bits back into words

>>> packed = VarUIntArray.packbits(bits, byteorder="big")
>>> packed
VarUIntArray([   1,    2, 1023], dtype='>u2', word_size=10)

Core Concepts

Word Size vs Machine Size

Standard computers work with word sizes of 8, 16, 32, or 64 bits. When you need a 10-bit word, it must be stored in a 16-bit container, leaving 6 padding bits unused. VarUIntArray automatically:

1. Selects the appropriate machine word size (8, 16, 32, or 64 bits)
2. Tracks the actual word size you care about
3. Ensures padding bits are handled correctly in operations

Byte Order

The byteorder parameter controls how multi-byte values are stored in memory. This matters when working with binary protocols or when interoperating with systems that expect a specific byte layout.

# Big-endian (most significant byte first)
>>> arr = VarUIntArray([0x0102], word_size=16, byteorder="big")
>>> arr.view(np.uint8)  # raw bytes: [0x01, 0x02]

# Little-endian (least significant byte first)
>>> arr = VarUIntArray([0x0102], word_size=16, byteorder="little")
>>> arr.view(np.uint8)  # raw bytes: [0x02, 0x01]

Byte order is preserved through slicing, ufuncs, concatenation, copy, serialization, and pack/unpack roundtrips. For single-byte word sizes (1-8 bits), byte order has no effect on storage but the attribute is still tracked.

Accepted values: 'big', 'little', 'native' (or NumPy prefixes '>', '<', '=').

Padding Bit Handling

The most important feature is correct handling of padding bits during bitwise operations. For example:

# 3-bit word stored in 8-bit container
>>> arr = VarUIntArray([5], word_size=3)  # Binary: 101

# Standard NumPy invert would give 11111010 (250)
# VarUIntArray.invert() gives 010 (2) - correct for 3-bit word
>>> inverted = arr.invert()
>>> int(inverted[0])
2

API Reference

VarUIntArray Class

Constructor

VarUIntArray(input_array, word_size, byteorder="native")

Parameters:

• input_array: Array-like data to convert
• word_size: Number of significant bits per word (1-64)
• byteorder: Byte order for the underlying dtype. Accepts 'big', 'little', 'native' or NumPy prefixes '>', '<', '='. Defaults to 'native'.

Methods

• invert(): Bitwise invert respecting word_size
• unpackbits(): Unpack to individual bits (adds one dimension)
• packbits(data, byteorder="native"): Class method to pack bit array into VarUIntArray
• to_dict(): Serialize to a dictionary
• from_dict(data): Static method to deserialize from a dictionary
• to_json(): Serialize to a JSON string
• from_json(string): Class method to deserialize from a JSON string

Attributes

• word_size: Number of significant bits per word
• byteorder: Byte order prefix ('>', '<', or '=')

Functions

unpackbits(array)

Unpack a VarUIntArray into individual bits, excluding padding.

>>> arr = VarUIntArray([5, 3], word_size=3)
>>> unpackbits(arr)
array([[1, 0, 1],
       [0, 1, 1]], dtype=uint8)

Parameters:

• array: VarUIntArray to unpack

Returns: ndarray with shape (*original_shape, word_size)

packbits(array, byteorder="native")

Pack a bit array into a VarUIntArray.

>>> bits = np.array([[1, 0, 1], [0, 1, 1]], dtype=np.uint8)
>>> packbits(bits)
VarUIntArray([5, 3], dtype=uint8, word_size=3)

Parameters:

• array: ndarray of uint8 containing 0s and 1s, where the last dimension contains bits for each word
• byteorder: Byte order for the result. Accepts 'big', 'little', 'native' or '>', '<', '='. Defaults to 'native'.

Returns: VarUIntArray with one fewer dimension

VarUIntArray.to_dict()

Serialize VarUIntArray to JSON-compatible dictionary. Includes byteorder as a human-readable string ('big', 'little', or 'native').

>>> arr = VarUIntArray([1, 2, 3], word_size=10, byteorder="big")
>>> arr.to_dict()
{'word_size': 10, 'byteorder': 'big', 'values': [1, 2, 3]}

VarUIntArray.from_dict(data)

Convert a dictionary to VarUIntArray. If byteorder is absent, defaults to 'native' for backwards compatibility.

>>> VarUIntArray.from_dict({'values': [1, 2, 3], 'word_size': 10, 'byteorder': 'big'})
VarUIntArray([1, 2, 3], dtype='>u2', word_size=10)

VarUIntArray.to_json()

Serialize VarUIntArray to a JSON string.

>>> arr = VarUIntArray([1, 2, 3], word_size=10, byteorder="big")
>>> arr.to_json()
'{"word_size": 10, "byteorder": "big", "values": [1, 2, 3]}'

VarUIntArray.from_json(string)

Deserialize a VarUIntArray from a JSON string.

>>> json_str = '{"word_size": 10, "byteorder": "big", "values": [1, 2, 3]}'
>>> VarUIntArray.from_json(json_str)
VarUIntArray([1, 2, 3], dtype='>u2', word_size=10)

Use Cases

Custom Binary Protocols

Working with network protocols or file formats that use non-standard bit widths:

# 12-bit color values (common in some image formats)
>>> colors = VarUIntArray([4095, 2048, 0], word_size=12)

Bit Manipulation

Performing bitwise operations on packed data:

>>> data = VarUIntArray([0b1010, 0b0101], word_size=4)
>>> mask = VarUIntArray([0b1100, 0b0011], word_size=4)
>>> result = data & mask  # Bitwise AND

Implementation Details

Memory Layout

• VarUIntArray defaults to machine-native byte order. Use the byteorder parameter to specify 'big', 'little', 'native' (or NumPy prefixes '>', '<', '=').
• Data is stored in the smallest standard NumPy unsigned integer type that can hold the specified word_size.

Limitations

• Maximum word size: 64 bits
• Only unsigned integers are supported
• The axis parameter is not supported for np.unpackbits on VarUIntArray

Examples

Complete Workflow

>>> import numpy as np
>>> from varuintarray import VarUIntArray

# Create some 5-bit values
>>> data = VarUIntArray([31, 16, 0, 15], word_size=5)

# Unpack to bits
>>> bits = data.unpackbits()
>>> bits
array([[1, 1, 1, 1, 1],
 [1, 0, 0, 0, 0],
 [0, 0, 0, 0, 0],
 [0, 1, 1, 1, 1]], dtype=uint8)

# Flip specific bits
>>> bits[:, 0] = 1 - bits[:, 0]  # Flip first bit

# Pack back
>>> result = VarUIntArray.packbits(bits)
>>> result
VarUIntArray([15, 0, 16, 31], dtype=uint8, word_size=5)

# Bitwise operations
>>> result.invert()
VarUIntArray([16, 31, 15,  0], dtype=uint8, word_size=5)

Serialization

>>> from varuintarray import VarUIntArray
>>> import json

# Serialize dict
>>> arr = VarUIntArray([100, 200, 300], word_size=12, byteorder="big")
>>> serialized = arr.to_dict()
>>> serialized
{'word_size': 12, 'byteorder': 'big', 'values': [100, 200, 300]}

# Deserialize dict
>>> VarUIntArray.from_dict(serialized)
VarUIntArray([100, 200, 300], dtype='>u2', word_size=12)

# Serialize JSON
>>> serialized = arr.to_json()
>>> serialized
'{"word_size": 12, "byteorder": "big", "values": [100, 200, 300]}'

# Deserialize JSON
>>> VarUIntArray.from_json(serialized)
VarUIntArray([100, 200, 300], dtype='>u2', word_size=12)

License

varuintarray is licensed under the MIT License - see the LICENSE file for details